Boehringer Ingelheim is the latest pharma company to turn to AI in the hunt for new treatments and therapies.
The German company announced earlier this week that it is working with IBM to use Big Blue’s foundation model tech “to discover novel candidate antibodies for the development of efficient therapeutics.”
The plan is to use an IBM-developed, pre-trained AI model and feed in additional proprietary data from Boehringer to speed up the discovery of potential antibodies and improve the quality of predicted candidates.
IBM’s biomedical foundation model relies on a wide range of public data sets, which include data on protein and drug target interactions. Mix in Boehringer’s proprietary data, and the hope is that newly designed proteins and small molecules with the desired properties will be generated.
Boehringer is not alone. Discovering and developing therapeutic antibodies – used in the treatment of diseases such as cancers, for example – is a time-consuming process. Novartis, for instance, has hooked up with Microsoft to apply AI technology in the hunt for new medicines. Pfizer is also using supercomputing and AI to get new treatments to patients faster.
And therein lies the rub.
While generative AI might be amusing when it comes to generating draft copy, kicking off some coding, or drawing pictures of people with worryingly wonky hands, careful thought is needed when the technology is used to develop candidates for therapies. It’s one thing to come up with a novel recipe for whisky. It’s quite different to use the tech where patient safety is concerned.
The pharmaceutical industry is notoriously conservative for good reason. Several regulatory bodies worldwide keep a close eye on their work to ensure that humans are not put at risk. Without putting too fine a point on it, the regulations governing the development and testing of treatments have been written in blood.
However, the pressure to speed up discovery continues to increase despite the tendency of AI services to suffer the odd hallucination or two. It’s not something you necessarily want to associate with pharmaceuticals.
We asked a number of regulatory agencies for their thoughts on how AI could be integrated into the therapeutic antibody discovery pipeline. The UK’s Medicines and Healthcare products Regulatory Agency (MHRA) promised a response but has yet to produce a comment.
The US’s Food and Drug Administration (FDA) told us it “intends to publish guidance over the next year with considerations on the use of AI in drug development.”
The agency added: “The guidance will provide high-level recommendations to sponsors who are considering the use of AI as part of producing information or data intended to support regulatory decision making for drugs.”
The European Medicines Agency (EMA) told us that while it provided reflections and guidance on regulations, it did not provide the regulations themselves.
That said, a spokesperson added: “Regarding the use of AI tools at different phases of a medicine’s lifecycle, marketing authorisation applicants (MAAs) or marketing authorisation holders (MAHs) who plan to deploy AI/Machine Learning (ML) technology are expected to consider and systematically manage relevant risks from early development to decommissioning.
“A key principle is that it is the responsibility of the MAA or MAH to ensure all algorithms, models, datasets and data processing pipelines used are fit for purpose and in line with ethical, technical, scientific and regulatory standards.”
The spokesperson noted that, from a regulatory agency perspective, applying AI in the drug discovery process may be a low risk setting, as the risk of non-optimal performance often mainly affects the sponsor.
“However, if results contribute to the total body of evidence which is presented for regulatory review, principles for non-clinical development should be followed. In this context, all models and datasets used would normally be reviewed by the sponsor.”
As for Boehringer Ingelheim itself, a spokesperson told The Register: “IBM’s foundation model used in this way would be a tool for scientific research and molecule design based on synthetic data.
“Only if it became involved in the final manufacture of medicines would Good Manufacturing Practices come into play, and only if the AI started to perform a definitive medical purpose (as per UK MDR 2002) would there be any need to consider this as AI as a Medical Device.” ®
